Package org.springframework.ai.openai.responses

Class OpenAiResponsesChatModel

java.lang.Object
org.springframework.ai.openai.responses.OpenAiResponsesChatModel
All Implemented Interfaces:
ChatModel, StreamingChatModel, Model<Prompt,ChatResponse>, StreamingModel<Prompt,ChatResponse>
public final class OpenAiResponsesChatModel extends Object implements ChatModel
A ChatModel backed by OpenAI's Responses API (/v1/responses), the endpoint where new OpenAI capability lands. Notably, it is the only way to combine reasoning with tool calling on GPT-5.4 and later, which Chat Completions no longer supports.

This model is stateless. Every call sends the whole Prompt; OpenAI is never asked to remember the conversation, so ChatMemory, advisors, RAG all keep working exactly as they do with OpenAiChatModel. The stateful modes are not supported.

Like OpenAiChatModel since 2.0.0, this model runs no tool loop: multi-round tool calling belongs to ToolCallingAdvisor, which is what keeps the behaviour identical across providers.

Known limitations: audio input and output, n, stop sequences, frequency and presence penalties, seed and logit_bias have no equivalent here; many OpenAI-compatible backends implement /v1/chat/completions but not /v1/responses; and reasoning continuity across turns depends on the chat memory repository in use, since none of them persists message parts yet - see the reference documentation for the details.

Since:
2.1.0
Author:
Dimitar Proynov
See Also: